Effects of manner and height on intraoral pressure and articulatory contact at voicing offset and onset for voiceless obstruentsa)

Laura L. Koenigb) Haskins Laboratories, 300 George Street, New Haven, Connecticut 06511 and Long Island University, One University Plaza, Brooklyn, New York 11201

Susanne Fuchs Center for General Linguistics (ZAS), Schuetzenstrasse 18, Berlin 10117, Germany

Jorge . Lucero Department of Mathematics, University of Brasilia, Brasilia DF 70910-900, Brazil (Received 13 September 2010; revised 3 February 2011; accepted 12 February 2011) In , a major constriction in the upper vocal tract yields an increase in intraoral pressure (Pio). Phonation requires that subglottal pressure (Psub) exceed Pio by a threshold value, so as the transglottal pressure reaches the threshold, phonation will cease. This work investigates how Pio levels at phonation offset and onset vary before and after different German voiceless (stop, , , clusters), and with following high vs low . Articulatory contacts, measured using electropalatography, were recorded simultaneously with Pio to clarify how supra- glottal constrictions affect Pio. Effects of consonant type on phonation thresholds could be explained mainly in terms of the magnitude and timing of vocal-fold abduction. Phonation offset occurred at lower values of Pio before fricative-initial sequences than stop-initial sequences, and onset occurred at higher levels of Pio following the unaspirated stops of clusters compared to frica- tives, affricates, and aspirated stops. The vowel effects were somewhat surprising: High vowels had an inhibitory effect at voicing offset (phonation ceasing at lower values of Pio) in short-duration consonant sequences, but a facilitating effect on phonation onset that was consistent across conso- nantal contexts. The vowel influences appear to reflect a combination of vocal-fold characteristics and vocal-tract impedance. VC 2011 Acoustical Society of America. [DOI: 10.1121/1.3561658] PACS number(s): 43.70.Aj, 43.70.Gr [DAB] Pages: 3233–3244

I. INTRODUCTION nected conditions that involve rapidly-varying glottal widths, upper vocal-tract postures, and intraoral pressures. Em- This paper investigates the effects of consonant manner pirical work on consonant voicing, in turn, has mostly assessed and following vowel height on the aerodynamic conditions the timing of phonation or the extent to which an obstruent at voicing offsets and onsets around voiceless obstruents and interval contains voicing (e.g., Docherty, 1992; Lisker and obstruent sequences. Phonation offsets and onsets were Abramson, 1964; Smith, 1997; Stevens et al.,1992; Weismer, determined for eight speakers of German, in six consonantal 1980; Westbury, 1979). Few studies have provided direct aero- contexts (/t/, /S/, clusters /st/, /St/, and affricates =tS=, =ts=) and ^ ^ dynamic and articulatory data about phonation offsets and two vowel contexts (following /A/ vs /i/ or /I/). At the times onsets in consonantal contexts, and those that have mostly of voicing offset and onset, intraoral pressure (P ) was io focused on voiced rather than voiceless obstruents. measured, as was lingua–palatal contact, using electropala- Thus, the general goal of the current work is to help tography (EPG). bridge this gap by quantifying the aerodynamic and supralar- Although voicing has been studied extensively, gaps yngeal conditions at the times of voicing offsets and onsets remain between the empirical data on obstruent voicing and before and after voiceless obstruent consonants produced by theoretical formulations of the physical requirements for pho- multiple speakers, and assessing the likely sources of varia- nation. As detailed in subsequent sections, modeling work has tion across consonants and vowels based on past work. generally focused on phonation thresholds in conditions with Along with providing a better understanding of the general adducted vocal folds and an open vocal tract. Results from relationships between speech articulation and aerodynamics, such studies cannot be straightforwardly extended to con- these data are intended to serve as input for aerodynamic modeling that incorporates vocal-fold abduction and varia- a)Portions of these data were presented at Acoustics 2008, Paris, France, and tions in supraglottal conditions. This paper will focus on the 2010 International Conference on Advances in Laryngeal Biophysiol- group patterns across consonant and vowel contexts. Future ogy/International Conference on Physiology and Biomechanics, work will explore speaker differences and relate the meas- Madison, WI. b)Author to whom correspondence should be addressed. Electronic mail: ured data with simulations obtained via modeling of the lar- [email protected] ynx and upper vocal tract.

J. Acoust. Soc. Am. 129 (5), May 2011 0001-4966/2011/129(5)/3233/12/$30.00 VC 2011 Acoustical Society of America 3233

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp A. Modeling of the factors that influence phonation offsets and onsets occur under different aerodynamic condi- tions. In particular, when laryngeal and supralaryngeal fac- 1. Phonation threshold pressures tors are held constant, a higher pressure is needed to initiate The physical requirements for voicing have been quanti- phonation than to maintain it, so that vocal-fold vibration fied for laryngeal models of varying complexity. Ishizaka shows hysteresis (Baer, 1975; Plant et al., 2004). Lucero and Matsudaira (1972) established that sustained phonation (1996, 1999, 2005) established the mathematical nature of in a two-mass model of the vocal folds requires a subglottal this phenomenon as a subcritical Hopf bifurcation and pressure exceeding a threshold value. Titze (1988) extended showed that hysteresis results from non-linear aerodynamic this line of inquiry to the body–cover model of the vocal damping within the vocal folds. In these models, the thresh- folds and provided the following formula for the phonation old pressure for maintaining phonation was found to be threshold pressure (Pth), about half for that initiating it. Studies in humans, however, suggest that the extent of hysteresis may vary considerably across speakers (Plant et al., 2004), presumably for the same Bcx0kt Pth ¼ ; (1) reasons as noted above for P variation. T th It should be noted that equivalent articulatory and aero- dynamic conditions do not hold at phonation offset and onset where B ¼ tissue damping, c ¼ the mucosal wave velocity in many consonantal contexts (see Sec. IB). Even if the (which increases with vocal-fold stiffness), x0 ¼ the glottal upper vocal tract is open, as for /h/ (cf. Koenig et al., 2005), half-width, kt ¼ a transglottal pressure coefficient calculated to onset–offset differences may not reflect true hysteresis be about 1.1, and T ¼ vocal fold thickness. Subsequent mathe- effects given variations related to syllable stress. The discus- matical work has explored how Pth varies with factors such as sion will consider how onset–offset differences in the current fundamental frequency (f0), prephonatory glottal width, glottal data compare to those in classic hysteresis studies in the light convergence angle, laryngeal size, and upper vocal-tract air- of such considerations. way inertance (Chan and Titze, 2006; Lucero, 1996, 1998, 2005; Lucero and Koenig, 2005, 2007; Titze, 1992). Many studies have also assessed Pth in human speakers, B. Assumptions required for characterizing phonation evaluating effects of vocal fatigue, vocal warm-up, hydra- around voiceless obstruents and review of empirical tion, and f0 (Chang and Karnell, 2004; Elliot et al., 1995; studies Finkelhor et al., 1987; Fisher et al., 2001; Jiang et al., 1999; Several assumptions were made in obtaining the phona- Milbraith and Solomon, 2003; Motel et al., 2003; Sivasankar tion threshold formula. First, the equation holds only for et al., 2008; Solomon and DiMattia, 2000; Solomon et al., cases where the vocal folds do not contact, i.e., when the 2003, 2007; Verdolini et al., 1994, 2002; Verdolini-Marston vocal folds are slightly abducted. The obstruent environ- et al., 1990). Several papers have reported extensive individ- ments considered here involve a wide range of vocal-fold ual differences in Pth values and changes across experimen- positions (i.e., glottal widths), from adducted to widely tal manipulations (Elliot et al., 1995; Finkelhor et al., 1987; abducted. Changes in laryngeal height during voiceless con- Jiang et al., 1999; Motel et al., 2003; Solomon et al., 2007; sonants may have passive effects on vocal-fold tension (e.g., Titze, 2009). The modeling work suggests that some of this Hombert et al., 1979), and speakers may also contract the variation may reflect speaker differences in vocal-fold pa- cricothyroid muscle to actively increase longitudinal tension rameters such as glottal convergence angle, resting glottal for voiceless consonants (Hoole and Honda, 2011; Lo¨fqvist half-width, tissue viscoelasticity, and thickness of the vocal- et al., 1989). Extensions of Titze’s model have demonstrated fold body and cover (Chan et al., 1997; Chan and Titze, that higher f0, as should result from greater vocal-fold ten- 2006; Lucero, 1998). sion, increases Pth (Lucero and Koenig, 2007). Thus, multi- Although most modeling has adopted values appropriate ple laryngeal factors may inhibit phonation around voiceless for an adult male , there are well-known laryngeal dif- consonants. ferences between men and women (Titze, 1989). Lucero and Further, most modeling work (and, accordingly, most Koenig (2005) observed that P was higher when a two- th work on Pth in humans) has considered the relatively open mass model was rescaled to values appropriate for women vocal-tract postures associated with vowel production. When (and children). Some work also suggests that phonatory the vocal tract is open, Pio remains low (close to atmospheric responses to conditions such as dehydration may differ pressure). The production of obstruents, in contrast, involves between men and women [cf. data from Solomon and the formation of an oral constriction and an increase in Pio. DiMattia (2000) vs Solomon et al. (2003) for women vs This pressure change, which is particularly rapid for conso- men, respectively]. Thus, questions remain about how fully nants that involve vocal-fold abduction, reduces the airflow existing models characterize the behavior of multiple speak- across the that fuels phonation. It may also lead to a ers, including women as well as men. more divergent glottal configuration and voice quality fea- tures characteristic of breathiness: Higher open quotients, more symmetrical waveshapes, and reduced vibratory ampli- 2. Offset–onset differences tudes (Bickley and Stevens 1986; see also Stevens, 1991; The Pth equation defines the requirements for initiating Titze, 2009). Hence, increases in Pio during obstruents may phonation. Many studies have demonstrated that phonation also have multiple inhibiting effects on phonation.

3234 J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp Finally, laryngeal conditions vary as a function of con- enhancement seems unlikely given the small magnitude of sonant manner. For an aspirated stop consonant, glottal area the effect (Weismer, 1980). remains fairly small as the upper vocal tract closes, whereas The preceding considerations suggest that a high vowel upon supraglottal release the vocal folds are abducted so as context could inhibit phonation, i.e., raise Pth. There are, to yield a long voicing delay (Lisker et al., 1970; Lo¨fqvist, however, alternative perspectives that lead to opposing pre- 1992; Lo¨fqvist and Yoshioka, 1981; Yoshioka et al., 1981). dictions. Titze (1988) noted that greater inertance of the air , on the other hand, seem to be more symmetrical column in the upper vocal tract facilitates phonation. Recent between offset and onset conditions, with peak glottal abduc- theoretical and modeling work (Lucero et al., 2009; Ruty tion occurring roughly in the middle of the frication noise et al., 2008) shows that increasing vocal-tract length (and (Hoole et al., 2003; Lo¨fqvist and Yoshioka, 1981, 1984; thereby lowering the first formant, F1) has the effect of Ridouane et al., 2006; Yoshioka et al., 1981). Speakers may reducing Pth, via increasing vocal-tract inertance. According abduct more extensively in fricatives than in stops (Hirose et to these results, lower Fls (characteristic of high vowels) al., 1978; Lindqvist, 1972; Lisker et al., 1969; Lo¨fqvist and should have a facilitating effect on phonation. Another line Yoshioka, 1981; Ridouane et al., 2006). Nevertheless, Pio of argument comes from studies that have investigated pho- rises to a higher peak value in stops than in fricatives, on av- nation for voiced . Ohala and Riordan (1979) and erage, given the presence of a complete occlusion in the for- Pape et al. (2006) observed that phonation may persist mer case vs a constriction (with some air escape) in the latter longer into voiced consonants preceding high vowels, evi- (Arkebauer et al., 1967; Fuchs and Koenig, 2009; Koenig dently as a function of larger pharyngeal cavity volume. et al., 1995). Similar effects could conceivably obtain for voiceless conso- In sum, given consonantal variations in the magnitude nants as well (cf. Mohr, 1971). and timing of laryngeal abduction, the buildup of Pio and secondary effects on laryngeal setting related to consonant D. Predictions production, one may expect that the aerodynamic conditions related to voicing offsets and onsets in obstruents will differ Based on the literature reviewed above, several predic- from what has been reported in studies assuming vocalic tions were made. conditions. 1. Onset–offset differences Since a higher threshold pressure is required to initiate C. Evidence for vowel effects on phonation phonation than to maintain it when all other conditions are held constant, one may generally expect that phonation Past work indicates that vowel postures may affect la- onsets will tend to occur at lower values of Pio than phona- ryngeal behavior in a variety of ways. One widely-studied tion offsets. (Note that higher Pio corresponds to lower trans- phenomenon is “intrinsic f0.” The essence of this effect is glottal pressure if Psub is constant; thus, factors that make that high vowels such as [i] and [u] have slightly higher f0’s, phonation more difficult to initiate or sustain should be asso- on average, than low vowels such as [A] (see Whalen and ciated with voicing at lower Pios). However, in the current Levitt, 1995, for an extensive cross- review). work, laryngeal, supralaryngeal, and aerodynamic conditions Although the precise origins of intrinsic f0 are not fully vary across phonetic contexts, so that many conditions are clear, it appears to result from elevated laryngeal height and/ not held constant. Thus (as detailed in the following para- or increased vocal-fold tension in high vowels. Vowel qual- graphs), it was expected that phonation offsets and onsets ity has also been shown to influence some aspects of voice would differ as a function of context. quality. The results of most relevance here are the following: Higgins et al. (1998) reported lower degrees of vocal-fold 2. Consonant differences contact in /i/ than /A/ and Chen et al. (2002) found lower speed quotients in /i/ compared to /A, ae, and u/. Such values Phonation offsets were predicted to occur at higher lev- consistent with higher breathiness in /i/ could indicate a glot- els of Pio for stops than fricatives. Closely-approximated tal setting less conducive to phonation in the high front vocal folds entering stops should facilitate phonation and vowels. allow it to persist to a higher level of Pio, whereas earlier Several studies have also found that the following vowel abduction for fricatives would inhibit phonation. One would affects voice onset times (VOTs; Lisker and Abramson, further expect that the first member in a consonantal 1964). Authors have reported longer VOTs before vowels sequence should have the greatest influence on voicing off- that are high (Chang, 1999; Docherty, 1992; Esposito, 2002; set, so that phonation offsets should be similar for /S/, /St/, = = =S= Higgins et al., 1998; Klatt, 1975; Ohala, 1981; Port and and /st/ on one hand and /t/, ^ts , and ^t on the other. Rotunno, 1979), long (Port and Rotunno, 1979), and/or tense For voicing onsets, a large glottal width at the release of (Weismer, 1979). The explanations that have been offered aspirated stops should inhibit phonation, so that voicing would for these findings are diverse: Higher supraglottal impedance resume at a low value of Pio. An important difference between in high vowels, leading to slower discharge of Pio (Chang, the single /t/ and that in clusters is that the latter is unaspirated 1999); elevated larynx position in high vowels (Klatt, 1975); in German. As such, phonation should resume at higher levels effects of stress on tense vs lax vowels (Weismer, 1979); of Pio after /st/ and /St/ compared to /t/. The earlier timing of or perceptual enhancement (Kingston, 1993). Perceptual abduction in fricatives compared to aspirated stops should

J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds 3235

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp also lead to phonation onset at higher levels of Pio for /S/, =tS=, TABLE I. Speech stimuli. Note that transcriptions in the table are phonetic, ^ and =ts= as compared to the aspirated stop. whereas the text uses the phonemic representations. ^ Finally, in =ts= vs =tS= and /st/ vs ^ ^ Consonant Following vowel Full word Gloss /St/) was not expected to yield large effects for offsets or onsets, but small effects might be observed owing to differ- [th][A] Tasche Bag ences in fricative channel width and tongue grooving [S][A] Schaf Sheep ½t S [A] Tschad Chad (the country) reported in past work (Dixit and Hoffman, 2007; Fletcher ^ ½ts [A] Zander Kind of fish and Newman, 1991; Recasens and Espinosa, 2007; Stone ^ [St] [A] Stachel Spike et al., 1992). [st] [A] Stalin Stalin h [t ][I] Tisch Table 3. Vowel effects [S][I] Schiff Ship ½t S [i] Tschibo Brand of coffee Several lines of evidence suggest that high vowels may ^ ½ts [I] Zitze Teat have an inhibitory effect on phonation, possibly because of ^ [St] [I] Stich Prick higher intrinsic f0 (which could relate to higher vocal-fold [st] [i] Stil Style stiffness), voice quality features associated with a breathy voicing posture, and/or higher impedance in a more con- stricted supraglottal tract. This would lead to phonation repetitions). The final data set (after removing productions with unanalyzable signals, disfluencies, or pauses) consisted occurring at lower values of Pio preceding high vowels com- pared to low ones. On the other hand, the source–filter inter- of 921 tokens. actions discussed in recent modeling work and effects of pharyngeal constrictions on pressure conditions near the B. Instrumentation and preliminary processing glottis could lead to precisely the opposite result. In either The Reading EPG System 3 was used to record lingua– case, vowel effects in the present data would presumably be palatal contacts, using a sampling rate of 100 Hz. A pressure stronger at voicing onset, adjacent to the target vowel, than transducer (Endevco 8507C-2, San Juan Capistrano, CA) was at voicing offset. affixed to the back end of the EPG palate via a small plastic tube and a separate tube was passed around the teeth to obtain II. METHODS a stable recording of atmospheric pressure. This recording The speakers and instrumentation used here were identi- method permits an assessment of Pio variation for sounds pro- cal to those described in Fuchs and Koenig (2009). The cur- duced anterior to the palatal place of articulation and is less rent study analyzed a larger set of utterances using different invasive than recording pharyngeal pressure via a catheter fed through the nose. The pressure signal was recorded to analysis methods (particularly for the Pio signal). PCQuirer using a sampling rate of 1859 Hz. An acoustic sig- nal was recorded to DAT at a sampling rate of 48 kHz. A. Speakers and stimuli In MATLAB, the Pio signals were smoothed using the filt- Eight adult speakers of Standard German were recorded: filt function and a kaiser window with 40 Hz passband and Three females (F1–F3) and five males (M1–M5). The conso- 100 Hz stopband edges, and a 50 dB damping factor. This nant and vowel contexts chosen for analysis are shown processing eliminated the high-frequency oscillation associ- in Table I. As indicated in the table, the target consonant ated with phonation and yielded minimal distortion around S =S= = = sequences were single /t/ and / /, the affricates ^t and ^ts , the regions of rapid pressure changes associated with obstru- and the clusters /St/ and /st/. The following vowel was either ent closure and release. The first derivative (velocity) of the 1 low (/A/) or high (/i/ or /I/). The single fricative was limited Pio signal was calculated from the smoothed signal. Voicing to /S/ because /s/ does not occur word-initially in German. It offsets and onsets were obtained interactively (i.e., visually) should be also noted that the cluster /st/ is rare in German, from the original (unsmoothed) Pio signal. Other measures occurring only as a variant in some borrowed words. Speak- described below were obtained automatically in MATLAB ers were instructed for the two words containing this cluster based on the times of voicing offsets and onsets. EPG frames (Stalin and Stil) to use the pronunciation /st/ rather than /St/. were also extracted at the times of voicing offsets and onsets, Finally, single /t/ is aspirated in German, whereas /t/ in the using the acoustic signal as a reference for alignment with clusters is unaspirated. the Pio. The words were produced in the sentential frame Ich To correct for drift in the Pio signal over the course of nasche ____ (“I nibble ____”), as the first member of a com- the recording session, a pressure minimum was obtained in pound with the word Stelle (“place”). Thus, the full utterance vowels on either side of the target consonant or consonantal for the first word was Ich nasche Taschenstelle. Initial posi- sequence. Initially, these minima were taken in the vowels tion in the compound put the target word in a prosodically- immediately before and after the target. In the course of focused position. Although the utterances used all real words obtaining these values, it was observed that the pressure of- of German, the sentences were mostly semantically nonsen- ten did not return to baseline when the target consonant pre- sical. Speakers produced each utterance ten times in a list ceded a high vowel (/i/ or /I/), not only in the high vowel with all utterances fully randomized. A total of 960 pro- itself, but also in the preceding schwa. Since one purpose of ductions was collected (8 speakers 12 utterances 10 this work was to quantify voicing thresholds in absolute

3236 J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp FIG. 1. Examples of selected fricative, stop, cluster, and productions as produced by speaker M4. From left to right: /@SA/, /@tA/, /@StA/, /@tSA/. ^ From top to bottom: Audio signal; original and smoothed Pio signals (where the smoothed is the heavy line); velocity of smoothed Pio. The circles in the Pio signals show locations of voicing offsets and onsets. The circles in the velocity signal show maximum positive and negative values. As described in the text, S S there were a few differences between the two clusters (/st/ and / t/) and the two affricates =^ts= and =^t = in the Pio values at phonation onset and offset, but the overall Pio profiles were qualitatively similar.

terms, for these high-vowel utterances, a baseline measure driving pressure could push the vocal folds into a stable vibra- was also obtained in the nearest low vowel, namely the /A/in tory pattern that would not be achieved with a more gradual nasche. This provided a value near the speaker’s baseline pressure change. The signals and measurement points are pressure (as evident between utterances) and was thus more shown for one production of a fricative, stop, cluster, and valid for present purposes. Finally, for all utterances, the affricate for one male speaker (M4) in Fig. 1.2 minimum pressure value in the neighboring vowels was To quantify tongue–palate contact patterns, two meas- determined (based on two vowels for utterances with /A/ ures were obtained for each EPG frame extracted at voicing words and three vowels for utterances with /i/ or /I/ words) offset and onset: The overall percentage of contact (PC) and and subtracted off the pressure values at voicing offset and the center of gravity (COG). The PC measure is simply the onset obtained from the smoothed Pio signal. total percentage of contacts on the EPG palate out of a possi- ble 62 (six electrodes in the most anterior row and eight elec- C. Measures trodes in the remaining seven rows). The COG measure weights contact in each row (Rl-8, numbered anterior to pos- After voicing offset and onset times had been determined, terior) by a coefficient that increases by 1, from 0.5 to 7.5, the following measures were obtained automatically in MAT- with anteriority. The sum of the weighted contacts is then di- LAB for each token: (a) The pressure values in the smoothed vided by the total number of contacts. Higher COG values Pio signal at voicing offset and offset, corrected for baseline correspond to more anterior places of articulation. drift as described above. (b) The peak positive pressure veloc- ity associated with the P rise into the consonant and the min- io D. Statistical methods imum negative pressure velocity associated with the Pio fall coming out of the consonant. This measure was included All statistics were computed in R (R Development Core following Lucero (2005), who found that a rapid increase in Team, 2008). To evaluate global differences between

J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds 3237

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp FIG. 2. Histogram of all pressure data (left) and box-plots of values for voicing onset and offset (right). In the box-plots, the thick lines in the box indicate the mean; the size of the boxes shows the 25–75th quartiles; and the dotted lines indicate the 10–90th percentiles. Isolated circles represent outliers.

phonation offsets and onsets, a paired t-test was calculated. tended to occur at higher values of Pio [Fig. 2 (right)]. The The vowel and consonant effects on Pio at voicing offsets group mean values were 350 Pa for voicing offsets vs 94 Pa and onsets were assessed by repeated-measures ANOVAs for voicing onsets. A two-sample t-test showed this differ- with vowel and consonant as independent factors and ence to be highly significant (t ¼ 54.35, p < 0.001, df ¼ 920). speaker and repetition as error terms. Pair-wise post-hoc At the same time, there is clearly a wide range of values tests were carried out to evaluate specific consonant effects. for both voicing offsets and onsets, as well as considerable For vowel effects, one-way ANOVAs were conducted for overlap between offset and onset values. The following sec- each consonantal context separately, with vowel height as tions explore how this variation, including high and low out- the independent factor and speaker and repetition as error lying values, reflects consonant and vowel context. The main terms. While it was of interest to assess whether there was effects from the repeated-measures ANOVA for vowel and an interaction between consonant and vowel effects in gen- consonant factors are given in Table II. Consonant and eral, a full set of pair-wise post-hoc analyses was not per- vowel height effects were significant for both phonation off- formed for all consonant–vowel combinations because not set and onset, as were the consonant-by-vowel interactions. all of these were deemed meaningful. For example, possible Results of post-hoc tests are provided below. differences in vowel effects for /tA/-/tI/vs/StA/-/StI/ were of interest, whereas differences between /tA/ and /StI/ were not. The EPG and pressure velocity data were reviewed in B. Differences across consonantal contexts selected cases to help interpret patterns in the pressure data. 1. Phonation offset

III. RESULTS It was predicted that phonation offsets would occur at SS S lower values of Pio for / t st/ as compared to =tt^ ts^=. Table A. Global Pio values at phonation offset and onset II (left) shows that the main effect of consonant on voicing offsets was significant; the post-hoc pair-wise tests are given Figure 2 (left) shows a histogram of all the Pio data pooled across speakers, consonants, and vowels. Figure 2 in Table III (left) and the data are plotted in Fig. 3. The SS (right) splits the data by phonation offset vs onset. It is evi- results were as expected: All comparisons between / t st/ vs =ttS ts= were significant and phonation ceased at lower val- dent that the pressure data were roughly bimodal distributed ^ ^ [Fig. 2 (left)] and that despite differences in laryngeal and ues of Pio before fricatives than stops. Two of the fricative- S S aerodynamic conditions across contexts, voicing offsets initial sequences, namely / / and / t/, account for most of the very low offset values seen in Fig. 2 (right). Differences among the stop and the affricates were not significant, nor

TABLE II. Main effects and interactions of the ANOVAs on Pio at voicing were differences among the single fricative and the clusters. offset and onset. Consonant (C) and vowel height (V height) were independ- ent factors; speaker and repetition were error terms. Values in boldface are significant at the 0.05 level. Post-hoc results are given in Table III for the 2. Phonation onset consonants and summarized in the text for the vowels. The main effect of consonant was also significant for Phonation offsets Phonation onsets voicing onsets (Table II, right). Post-hoc results are given in Factor DFs F-value p-value Factor DFs F-value p-value Table III (right) and the data are graphed in Fig. 4. The pri- mary expectation was that aspirated vs unaspirated stops C 5 98.120 <0.001 C 5 136.848 <0.001 would differ. The data were consistent with this prediction: V height 1 8.392 0.004 V height 1 408.066 <0.001 C V height 5 2.283 0.045 C V height 5 12.641 <0.001 Phonation began at higher values of Pio after unaspirated Residuals 894 Residuals 894 stops (i.e., those in the clusters) than the aspirated ones. The clusters account for most of the high-end outliers seen for

3238 J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp TABLE III. Post-hoc results (p-values) for Pio at phonation offset and onset as a function of consonant. For 15 pair-wise comparisons at each location, the adjusted a is 0.00333. Significant comparisons are indicated in boldface.

Phonation offsets Phonation onsets

SS S SS S tstt^ ts ^t tstt^ tst^ S – 0.195 0.092 <0.001 <0.001 <0.001 S – <0.001 <0.001 0.020 <0.001 0.009 St – 0.016 <0.001 <0.001 <0.001 St – 0.033 <0.001 <0.001 <0.001 st – <0.001 <0.001 <0.001 st – <0.001 <0.001 <0.001 t – 0.157 0.347 t – <0.001 <0.001 ts – 0.467 ts – < ^ ^ 0.001

onsets in Fig. 2 (right). Pio at phonation onset for the unaspi- 1. Phonation offset rated stops was also significantly higher than for the fricative Data for voicing offsets as a function of vowel are given and affricates. Qualitatively, the aspirated stop showed the in Fig. 6. The post-hoc tests showed that, of the six conso- lowest P at phonation onset of all the contexts; the differ- io nantal contexts, vowel effects were significant (p < 0.001, ence was not significant for the single fricative, but it was using an adjusted a of 0.05/6 ¼ 0.00833) for the single stop, for the affricates. the single fricative, and the cluster /St/, with phonation per- A difference was also observed between the two sisting to higher levels of P before low vowels. This affricates: Phonation began at slightly higher values of P io io implies that a higher transglottal pressure was needed to following =ts= than =tS=. This might be a result of more ^ ^ maintain phonation before a high vowel. articulatory contact for =tS= (i.e., greater supraglottal con- ^ The significant vowel effects for the singleton conso- striction). The EPG data showed that, for most speakers, av- nants can be explained by their shorter duration as compared erage PC values at voicing onset were higher following the to the clusters and affricates (cf. the durational data for a postalveolar than the alveolar. The overall difference was subset of German stops, fricatives, affricates, and clusters in very small, however (about 3%), and reflected a pattern Fuchs and Koenig, 2009). That is, a shorter transvocalic observed in the high vowel context. Specifically, seven out consonant interval permits more influence of the following of the eight speakers had higher average PC values for =tSi= ^ vowel. Durational effects also appear to explain the signifi- than =tsI= (group average of 10%). Conversely, seven out of ^ cant vowel effect for /St/ but not /st/: The cross-speaker aver- eight had the reverse EPG pattern in the low vowel context age duration of /St/, across the two vowel contexts, was 201 (more contact for /tsA/ than /tSA/), but in this case the aver- ^ ^ ms (SD ¼ 49 ms), whereas for /st/ it was 237 ms (SD ¼ 59 age difference was only 3%. To help clarify these PC values, ms). This pattern was quite consistent across speakers and EPG plots, showing contact frequency over all repetitions, may reflect the fact that /st/ is rare in German (i.e., speakers are shown for a representative speaker (M5) in Fig. 5. This may have tended to produce this cluster more slowly and particular speaker had PC values of 33.4% for =tSi= vs ^ carefully). The vowel effect for /St/ but not /st/ does not 24.8% for =tsI= (8.5% difference), compared to 18.3% for ^ appear to reflect differences in the degree or placement of /tSA/ and 19.4% for /tsA/ (1.1% difference). ^ ^ articulatory contact. Vowel-related differences in average PC and COG values at voicing offset were very small and C. Differences across vowel contexts virtually identical for the two clusters: For PC, a vowel dif- The main effect of vowel was significant for both voic- ference of 0.71% for /St/ vs 1.18% for /st/; and for COG, ing offsets and onsets, as was the consonant-by-vowel inter- 0.04 for /St/ vs 0.25 for /st/. action (Table II). Thus, the plots in this section show the data split by both consonant and vowel. However, as 2. Phonation onset explained in Sec. II D, the statistics only assessed vowel effects within each consonantal context. The pattern for phonation onsets was more consistent: For all consonantal contexts, voicing onsets occurred at

FIG. 3. Pio at phonation offsets as a function of consonant. FIG. 4. Pio at phonation onsets as a function of consonant.

J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds 3239

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp FIG. 5. EPG frequency plots for all utterances of one speaker. Each plot shows electrode rows 1–8, from top to bottom; row 1, at the top, is the most anterior. Within the frequency plot for each utterance, black cells indicate contact in 76%–100% of productions; dark gray in 51%–75% of productions; light gray in 26%– 50% of productions; unfilled in 0%– 25% of productions.

significantly higher levels of Pio before high vowels be attributed to three of the eight speakers, who showed lon- (p < 0.001, with a ¼ 0.00833). The effect was also consistent ger VOTs before /A/ on the order of 10 ms. The other five across speakers. The group data are given in Fig. 7, and Fig. speakers either had equivalent VOTs for the two vowel 8 shows one representative pair in the /t/ context for contexts, or very slightly longer VOTs in the high vowel speaker Fl. context (<10 ms). Because of these speaker differences, the One reason vowel effects were of interest here was ANOVA was also run without speakers as an error term. In because of past studies reporting longer VOTs before high this case, the p-value was 0.258. We conclude that vowel vowels. The current data did not show such an effect. An context did not consistently affect VOT in these data. ANOVA on VOT measures for /t/ (vowel height as the inde- Despite the lack of a clear vowel effect on VOT in the pendent variable, speaker and repetition as error terms) expected direction, other vowel-related effects postulated by yielded a p-value of 0.029, with VOTs being slightly longer past authors could still hold, not only stops but also across before the low vowel (about 3 ms on average). A review of consonantal contexts. As outlined in Sec. IC, the most individual speakers indicated that this average effect could common considerations have included effects of (a) vowel

3240 J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp FIG. 7. Pio at phonation onsets as a function of vowel (h ¼ high, l ¼ low) in FIG. 6. Pio at phonation offsets as a function of vowel (h ¼ high, l ¼ low) in different consonantal contexts. different consonantal contexts.

articulation on vocal-fold characteristics, especially tension differences observed here must be ascribed to these f0 differ- (leading to the intrinsic f0 effect), and (b) supralaryngeal ences (Lucero, 2005; Lucero and Koenig, 2007; Lucero et constriction degree on the rate of Pio discharge. To evaluate al., 2011; Solomon et al., 2007; Verdolini-Marston et al., whether these hypothesized conditions held here, follow-up 1990), as well as the consonant and vowel effects discussed analyses were carried out on the EPG and pressure velocity below. These contextual factors provide an explanation for data, as well as f0 (measured from the acoustic signal over why the offset–onset differences found here are larger than the first three pulses of the post-consonantal vowel). The those observed in past studies. In these data (see Fig. 2 and intrinsic f0 effect was evident in the data: In the high vowel associated discussion), phonation onset occurred, on aver- context, the average f0 across speakers was 177.8 Hz age, at 94 Pa vs 350 Pa for offsets, a difference of 256 Pa (SD ¼ 54.7 Hz), compared to 152.1 Hz (SD ¼ 45.95 Hz) in and a ratio of 3.73. For comparison, Lucero (2005) obtained the low vowel context, with males and females showing a a difference of about 120 Pa for a two-mass model of the comparable frequency ratio. The average PC values from the vocal folds and Lucero (1995) reported a ratio of about 2 for EPG data also showed the predicted greater contact in the the body–cover model of the vocal folds [cf. also compara- high vowel context (across speakers and consonantal con- ble data from Baer (1975) on canine larynges and Titze et al. texts, a difference of about 12%). PC data for all consonants, (1995) on a physical model]. Thus, the direction of the cur- split by vowel, are shown in Fig. 9. This difference was, rent onset–offset differences is consistent with past modeling moreover, reflected in the rate of the pressure change at work, but the magnitude of the effect is considerably larger release (as seen in the example in Fig. 8). The velocity mini- and could not have been easily predicted. mum reflecting the Pio discharge averaged -9.89 Pa/s for the It should also be noted that the current study used Pio high vowel context (SD ¼ 5.70), compared to -14.71 Pa/s for data to infer transglottal pressure, assuming that Psub was con- the low vowels (SD ¼ 8.53). The direction was consistent stant. Several considerations limit the accuracy of this across consonantal contexts, but the magnitude was most assumption. First, Psub may decrease a bit when the vocal extreme in the clusters (where the unaspirated stops were associated with a very rapid Pio decrease; cf. Fig. 1). In short, predictions of past work based on intrinsic f0 and rate of pressure discharge in high vowel contexts were met, yet still the high vowel context had a facilitating effect on pho- nation onsets.

IV. DISCUSSION As expected, on average phonation onsets occurred at lower values of Pio than phonation offsets. As reviewed in Sec. IA2, past studies have established that, under identical conditions, initiating vocal-fold vibration requires a higher level of transglottal pressure than sustaining it (i.e., there is hysteresis). In this study, onset–offset differences are not appropriately referred to as hysteresis, however, since the conditions at offset and onset are not symmetrical. Differen- ces arise as a function of supraglottal conditions, laryngeal– supralaryngeal timing, and vocal-fold characteristics related to stress and accent. Since the target consonant here initiated a stressed syllable and the word was in a prosodically- FIG. 8. Unsmoothed Pio data for two individual productions of /tA/ (gray) and /tI/ (black) from speaker Fl. These examples show the typical pattern of focused position, f0 was generally higher on the second phonation onset occurring at a higher value of Pio for the high vowel vowel of the sequence. Some component of the offset–onset context.

J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds 3241

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp B. Vowel effects The vowel effects on phonation offset and onset present a paradox. At least for the shorter consonantal sequences, following high vowels had a slight inhibitory effect on pho- nation, with voicing offsets occurring at lower values of Pio. At voicing onset, a very consistent effect in the opposite direction was observed: Phonation began at higher levels of Pio entering high vowels. It was predicted that the following vowel context would have a greater influence on phonation onsets than on phonation offsets, and comparison of Fig. 6 and 7 shows that vowel effects were much larger in the case of phonation onset. The most sensible explanation for the inhibitory effect FIG. 9. EPG percent contact (PC) at phonation onset as a function of conso- of high vowels on phonation offsets would appear to be an nant and vowel (h ¼ high, l ¼ low). anticipatory laryngeal raising for a following high vowel, leading to slight increases in vocal-fold tension. In contrast, folds are widely abducted (Hirose and Niimi, 1987; Lucero the effects for voicing onsets go against expectations based and Koenig, 2005). Our speakers also used self-selected loud- on intrinsic f0 (whereby more vocal-fold tension in the high ness and pitch, and variations in these factors will contribute vowel context should inhibit phonation), as well as those some noise to the data. For the most part, any such variations based on supraglottal areas (whereby slower Pio discharge (as well as any vocal fatigue effects over the course of the re- during the release phase of obstruents before high vowels cording session) should be randomly distributed across utter- should inhibit phonation). The explanation that does seem to ances, although the words containing the rare cluster /st/ may account for these data is one based on vocal-tract inertance have been produced with systematically greater loudness and (Lucero et al., 2009; Ruty et al., 2008). As indicated in Sec. pitch as well as the observed greater duration. A post-hoc IC, these studies have demonstrated reduced Pth values with analysis indicated that the /st/ clusters had slightly higher av- lower values of F1. erage peak pressures within the consonantal interval on the order of 40 Pa. Data from other will be needed to V. CONCLUSIONS clarify the extent to which the small voicing differences The consonant and vowel patterns in the current data S observed between /st/ and / t/ here reflect the effects of rarity reflect the multitude of factors that affect phonation thresh- vs intrinsic differences in supraglottal articulation and aerody- olds and contribute to an understanding of the nature and namics between the two sequences. Modeling can also help magnitude of supraglottal influences on vocal fold behavior. disentangle the influences of all of these factors. Although voicing offsets, on average, occurred at higher val- ues of P than voicing onsets, there was considerable over- A. Consonant effects io lap between onset and offset values and contextual effects Reports of earlier glottal abduction for fricatives com- on onset–offset patterns were substantial. The Pio levels at pared to stops led to the prediction that voicing would cease phonation onset and offset across consonantal contexts could at lower values of Pio before fricatives and fricative-initial generally be explained in terms of differing supraglottal con- sequences (i.e., clusters) as compared to stops and stop-ini- ditions, as evident in the EPG data, and laryngeal-oral tim- tial sequences (i.e., affricates). The data bore out this expec- ing. The Pio patterns for the vowels were more complex than tation. For phonation onsets, the main pattern was that those for the consonants. For the current data, explanations phonation began at higher levels of Pio for the unaspirated for the opposing onset vs offset patterns could be found in stops of the clusters compared to the other contexts. It was past studies of vowel effects on laryngeal states and recent somewhat surprising that phonation onset was not different work on the influence of vocal-tract inertance on Pth, but for the aspirated stop vs the single fricative, since the maxi- given the limited set of vowel contrasts used here, more mum abduction in the aspirated stop occurs very late relative work is needed to clarify how adjacent vowels affect phona- to supraglottal closure compared to the fricative (Hoole tion onsets and offsets around consonants. et al., 2003; Lofqvist and Yoshioka, 1981, 1984; Ridouane The methodology used here was designed specifically to et al., 2006; Yoshioka et al., 1981). One possibility is that a permit non-invasive recording of naturalistic speech from faster pressure drop at release compared to fricative multiple speakers. Assessing the adequacy of laryngeal and release (cf. examples in Fig. 1) helps counteract the wide supralaryngeal modeling to describe typical speech patterns glottal aperture at aspirated stop release. Lucero (2005) requires such data. At the same time, the present method left observed that a rapid pressure change could be more effec- some factors uncontrolled (e.g., loudness, con- tive at initiating phonation than a gradual change. Phonation tours, and the tense–lax feature of vowels). Clarifying the at high levels of Pio after the unaspirated stops could then effects of these parameters will require focused empirical reflect the combination of the adducted glottal posture and studies that exercise control over specific variables of inter- the rapid pressure change. Both of these factors contribute to est, as well as modeling work that permits changing single early phonation onsets. parameters independently.

3242 J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp ACKNOWLEDGMENTS Fuchs, S., and Koenig, L. L. (2009 ). “Simultaneous measures of electropala- tography and intraoral pressure in selected voiceless lingual consonants We express our appreciation to our speakers; to Jo¨rg and consonant sequences of German,” J. Acoust. Soc. Am. 126, 1988– Dreyer for designing the experimental setup; and to Anders 2001. Lo¨fqvist, Doug Whalen, and two anonymous reviewers for Higgins, M. B., Netsell, R., and Schulte, L. (1998). “Vowel-related differen- ces in laryngeal articulatory and phonatory function,” J. Speech Lang. comments on earlier versions of this manuscript. This study Hear. Res. 41, 712–724. was funded by a grant to the PILIOS project from the DFH Hirose, H., and Niimi, S. (1987). “The relationship between glottal opening Saarbru¨cken, a grant from the Federal Ministry of Education and the transglottal pressure differences during consonant production,” in Laryngeal Function in Phonation and Respiration, edited by T. Baer, C. and Research (BMBF) to PB1 at ZAS, and grants from Sasaki, and K.S. Harris (College-Hill Press, Boston, MA), pp. 381–390. MCT/CNPq (Ministe´rio da Cieˆncia e Tecnologia/Conselho Hirose, H., Yoshioka, H., and Niimi, S. (1978). “A cross language study of Nacional de Desenvolvimento Cientı´fico e Tecnolo´gico) and laryngeal adjustment in consonant production,” Annu. Bull. Res. Inst. CAPES (Coordenac¸a˜o de Aperfeic¸oamento de Pessoal de Logoped. Phoniatr. Univ. Tokyo 12, 61–71. Hombert, J.-M., Ohala, J. J., and Ewan, W. G. (1979). “Phonetic explana- Nı´vel Superior), Brazil. tions for the development of tones,” Language 55, 37–58. Hoole, P., and Honda, K. (2011). “Automaticity vs feature-enhancement in the control of segmental F0,” in Where do phonological contrasts come 1It is acknowledged that the vowel set does not control for tense vs lax from? Cognitive, physical and developmental bases of phonological quality. This is true for /i/ vs /I/ as well as for /A/ in different syllabic con- features, edited by N. Clements and R. Ridouane (John Benjamins, texts: The words Schaf and Stachel contain the lax variant of the vowel. Amsterdam), pp. 131–171. As such, the vowel context was coded as one of height only. It was not Hoole, P., Fuchs, S., and Dahlmeier, K. (2003). “Interarticulator timing in possible to fully control for tense–lax and still use real words of German. initial consonant clusters,” in Proceedings of the Sixth International Semi- One goal of the work was to analyze fairly naturalistic speech, so we opted nar on , edited by S. Palethorpe and M. Tabain (Mac- for real words recognizing that this would entail some loss of control over quarie University, Sydney), pp. 101–106. all phonetic features for the vowels. Ishizaka, K., and Matsudaira, M. (1972). Fluid Mechanical Considerations 2As evident in Fig. 1, the pressure velocity signals sometimes had multiple of Vocal Cord Vibration, Speech Communication Research Laboratory positive or negative excursions entering or exiting the consonant. For pres- Monograph No. 8, (Speech Communications Research Laboratory, Santa ent purposes, a single positive/negative pressure velocity value was Barbara, CA). obtained. However, in future work it may be of interest to characterize Jiang, J., Hanson, D. G., Conley, D., and O’Mara, T. (1999). “Phonation threshold pressure measurements during phonation by airflow inter- such tokens in order to more fully understand the Pio changes associated with obstruent production. ruption,” Laryngoscope 109, 425–432. Kingston, J. (1993). “The and of perceptually moti- vated articulatory covariation,” Lang. Speech 35, 99–113. Arkebauer, H. J., Hixon, T. J., and Hardy, J. C. (1967). “Peak intraoral air Klatt, D. H. (1975). “, frication, and aspiration in word-ini- pressures during speech,” J. Speech Hear. Res. 10, 196–208. tial consonant clusters,” J. Speech Hear. Res. 18, 686–706. Baer, T. (1975). “Investigation of phonation using excised larynges,” Ph. D. Koenig, L. L., Lofqvist, A., Gracco, V. L., and McGowan, R. S. (1995). dissertation, MIT, Cambridge, MA. “Articulatory activity and aerodynamic variation during voiceless conso- Bickley, C. A., and Stevens, K. N. (1986). “Effects of a vocal-tract constric- nant production,” J. Acoust. Soc. Am. 97, 3401. tion on the glottal source: Experimental and modelling studies,” J. Pho- Koenig, L. L., Mencl, W. E., and Lucero, J. C. (2005). “Multidimensional netics 14, 373–382. analyses of voicing offsets and onsets in female speakers,” J. Acoust. Soc. Chan, R. W., and Titze, I. R. (2006). “Dependence of phonation threshold Am. 118, 2535–2550. pressure on vocal tract acoustics and vocal fold tissue mechanics,” J. Lindqvist, J. (1972). Laryngeal Articulation Studied on Swedish Subjects, Acoust. Soc. Am. 119, 2351–2362. Speech, Music and Hearing Quarterly Progress and Status Report (Royal Chan, R. W., Titze, I. R., and Titze, M. R. (1997). “Further studies of phona- Institute of Technology, KTH, Stockholm, Sweden), pp. 2–3, 10–27. tion threshold pressure in a physical model of the vocal fold mucosa,” J. Lisker, L., and Abramson, A. S. (1964). “A cross-language study of voicing Acoust. Soc. Am. 101, 3722–3727. in initial stops: Acoustical measurements,” Word 20, 384–422. Chang, A., and Karnell, M. P. (2004). “Perceived phonatory effort and pho- Lisker, L., Abramson, A. S., Cooper, F. S., and Schvey, M. H. (1969). nation threshold pressure across a prolonged voice loading task: A study “Transillumination of the larynx in running speech,” J. Acoust. Soc. Am. of vocal fatigue,” J. Voice 18, 454–466. 45, 1544–1546. Chang, S. S. (1999). “Vowel dependent VOT variation,” in Proceedings of Lisker, L., Sawashima, M., Abramson, A. S., and Cooper, F. S. (1970). the 14th International Congress of Phonetic Sciences, edited by J. J. Cinegraphic Observations of the Larynx During Voiced and Voiceless Ohala, Y. Hasegawa, M. Ohala, D. Graville and A. C. Bailey (ICPhS, Stops, Haskins Laboratories Status Reports on Speech Research SR-21/22, Berkeley), pp. 1021–1024. Haskins Laboratories, New Haven, CT, pp. 201–210. Chen, Y., Robb, M. P., and Gilbert, H. R. (2002). “Electroglottographic Lo¨fqvist, A. (1992). “Acoustic and aerodynamic effects of interarticulator evaluation of gender and vowel effects during modal and fry phonation,” timing in voiceless consonants,” Lang. Speech 35, 15–28. J. Speech Lang. Hear. Res. 45, 821–829. Lo¨fqvist, A., Baer, T., McGarr, N. S., and Story, R. S. (1989). “The crico- Dixit, R. P., and Hoffman, P. R. (2007).“Articulatory characteristics of frica- thyroid muscle in voicing control,” J. Acoust. Soc. Am. 85, 1314–1321. tives and affricates in Hindi: An electropalatographic study,” J. Int. Pho- Lo¨fqvist, A., and Yoshioka, H. (1981). “Interarticulator programming in netic Assoc. 34, 141–159. obstruent production,” Phonetica 38, 21–34. Docherty, G. (1992). The Timing of Voicing in British English Obstruents Lo¨fqvist, A., and Yoshioka, H. (1984). “Intrasegmental timing: Laryngeal- (Foris, Berlin), pp. 1–289. oral coordination in voiceless consonant production,” Speech Commun. 3, Elliot, N., Gramming, P., and Sundberg, J. (1995). “What happens during 279–289. vocal warm up?” J. Voice 9, 37–44. Lucero, J. C. (1995). “The minimum lung pressure to sustain vocal fold Esposito, A. (2002). “On vowel height and consonantal voicing effects: oscillation,” J. Acoust. Soc. Am. 98, 779–784. Data from Italian,” Phonetica 59, 197–231. Lucero, J. C. (1996). “Relation between the phonation threshold pressure Finkelhor, B. K., Titze, I. R., and Durham, P. L. (1987). “The effect of vis- and the prephonatory glottal width in a rectangular glottis,” J. Acoust. cosity changes in the vocal folds on the range of oscillation,” J. Voice 1, Soc. Am. 100, 2551–2554. 320–325. Lucero, J. C. (1998). “Optimal glottal configuration for ease of phonation,” Fisher, K. V., Ligon, J., Sobecks, J. L., and Roxe, D. M. (2001). “Phonatory J. Voice 12, 151–158. effects of body fluid removal,” J. Speech Lang. Hear. Res. 44, 354–367. Lucero, J. C. (1999). “A theoretical study of the hysteresis phenomenon at Fletcher, S. G., and Newman, D. G. (1991). “[s] and [S] as a function of lin- vocal fold oscillation onset–offset,” J. Acoust. Soc. Am. 105, 423–431. guapalatal contact place and groove width,” J. Acoust. Soc. Am. Lucero, J. C. (2005). “Oscillation hysteresis in a two-mass model of the 89, 850–858. vocal folds,” J. Sound Vib. 282, 1247–1254.

J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds 3243

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp Lucero, J. C., and Koenig, L. L. (2005). “Phonation thresholds as a function impaired compensation to oral breathing in individuals with a history of of laryngeal size in a two-mass model of the vocal folds,” J. Acoust. Soc. vocal fatigue,” J. Speech Hear. Res. 51, 1494–1506. Am. 118, 2798–2801. (Letter) Smith, C. (1997). “The devoicing of /z/ in American English: Effects of Lucero, J. C., and Koenig, L. L. (2007). “On the relation between the phona- local and prosodic context,” J. Phonetics, 25, 471–500. tion threshold lung pressure and the oscillation frequency of the vocal Solomon, N. P., and DiMattia, M. S. (2000). “Effects of a vocally fatiguing folds,” J. Acoust. Soc. Am. 121, 3280–3283. task and systemic hydration on phonation threshold pressure,” J. Voice 14, Lucero, J. C., Van Hirtum, A., Ruty, N., Cisonni, J., and Pelorson, X. 341–362. (2009). “Validation of theoretical models of phonation threshold pressure Solomon, N. P., Glaze, L. E., Arnold, R., and Van Mersbergen, M. (2003). with data from a vocal fold mechanical replica,” J. Acoust. Soc. Am. 125, “Vocal fatigue and systemic hydration in men,” J. Voice 17, 31–46. 632–635. (Letter) Solomon, N. P., Ramanathan, P., and Makashay, M. J. (2007). “Phonation Lucero, J. C, Koenig, L. L., Lourenc¸o, K. G., Ruty, N., and Pelorson, X. threshold pressure across the pitch range: Preliminary test of a model,” J. (2011). “A lumped mucosal wave model of the vocal folds revisited: Voice 21, 541–550. Recent extensions and oscillation hysteresis,” J. Acoust. Soc. Am. 129, Stevens, K. N. (1991). “Vocal-fold vibration for obstruent consonants,” in 1568–1579. Vocal Fold Physiology: Acoustic, Perceptual, and Physiological Aspects Milbraith, R. L., and Solomon, N. P. (2003). “Do vocal warm-up exercises of Voice Mechanisms, edited by J. Gauffin and B. Hammarberg (Singular alleviate vocal fatigue?” J. Speech Lang. Hear. Res. 46, 422–436. Publishing Group, Inc., San Diego), pp. 29–36. Mohr, B. (1971). “Intrinsic variations in the speech signal,” Phonetica 23, Stevens, K. N., Blumstein, S. E., Glicksman, L., Burton, M., and Kurowski, 65–93. K. (1992). “Acoustic and perceptual characteristics of voicing in fricatives Motel, T., Fisher, K. V., and Leydon, C. (2003). “Vocal warm-up increases and fricative clusters,” J. Acoust. Soc. Am., 91, 2979–3000. phonation threshold pressure in soprano singers at high pitch,” J. Voice Stone, M., Faber, A., Raphael, L. J., and Shawker, T. H. (1992). “Cross- 17, 160–167. sectional and linguopalatal contact patterns in [s], [f], and Ohala, J. J. (1981). “Articulatory constraints on the cognitive representation [1],” J. Phonetics 20, 253–270. of speech,” in The Cognitive Representation of Speech, edited by T. Myers, Titze, I. R. (1988). “The physics of small-amplitude oscillation of the vocal J. Laver and J. Anderson (North-Holland, Amsterdam), pp. 111–122. folds,” J. Acoust. Soc. Am. 83, 1536–1552. Ohala, J. J., and Riordan, C. J. (1979). “Passive vocal tract enlargement dur- Titze, I. R. (1989). “Physiologic and acoustic differences between male and ing voiced stops,” in Speech Communication Papers Presented at the 97th female voices,” J. Acoust. Soc. Am. 85, 1699–1707. Meeting of the Acoustical Society of America, edited by Jared J. Wolf and Titze, I. R. (1992). “Phonation threshold pressure: A missing link in glottal Dennis H. Klatt (MIT Press, Cambridge, MA), p. 89. aerodynamics,” J. Acoust. Soc. Am. 91, 2926–2935. Pape, D., Mooshammer, C., Hoole, P., and Fuchs, S. (2006). “Devoicing of Titze, I. R. (2009). “Phonation threshold pressure measurement with a semi- word-initial stops: A consequence of the following vowel?” in Speech occluded vocal tract,” J. Speech Lang. Hear. Res. 52, 1062–1072. Production: Models, Phonetic Processes, and Techniques, edited by J. Titze, I. R., Schmidt, S. S., and Titze, M. R. (1995). “Phonation threshold Harrington and M. Tabain (Psychology, New York), pp. 211–226. pressure in a physical model of the vocal fold mucosa,” J. Acoust. Soc. Plant, R. L., Freed, G. L., and Plant, R. E. (2004). “Direct measurement of Am. 97, 3080–3084. onset and offset phonation threshold pressure in normal subjects,” J. Verdolini-Marston, K., Drucker, D. G., and Titze, I. R. (1990). “Changes in Acoust. Soc. Am. 116, 3640–3646. phonation threshold pressure with induced conditions of hydration,” J. Port, R. F., and Rotunno, R. (1979). “Relation between voice-onset time and Voice 4, 142–151. vowel duration,” J. Acoust. Soc. Am. 66, 654–661. Verdolini, K., Min, Y., Titze, I., Lemke, J., Brown, K., van Mersbergen, R Development Core Team (2008). “R: A language and environment for sta- M., Jiang, J. J., and Fisher, K. (2002). “Biological mechanisms underly- tistical computing. [R Foundation for Statistical Computing, Vienna, Aus- ing voice changes due to dehydration,” J. Speech Lang. Hear. Res. 45, tria: ISBN 3-900051-07-0],” http://www.R-project.org (Last viewed 268–281. October 20, 2008). Verdolini, K., Titze, I. R., and Fennell, A. (1994). “Dependence of phona- Recasens, D., and Espinosa, A. (2007). “An electropalatographic and acous- tory effort on hydration level,” J. Speech Hear. Res. 37, 1001–1007. tic study of affricates and fricatives in two Catalan dialects,” J. Int. Pho- Weismer, G. (1979). “Sensitivity of voice-onset time (VOT) measures to netic Assoc. 37, 143–172. certain segmental features in speech production,” J. Phonetics 7, 197–204. Ridouane, R., Fuchs, S., and Hoole, P. (2006). “Laryngeal adjustments Weismer, G. (1980). “Control of the voicing distinction for intervocalic in the production of voiceless obstruent clusters in Berber,” in Speech stops and fricatives: Some data and theoretical considerations,” J. Pho- Production: Models, Phonetic Processes, and Techniques, edited by netics, 8, 427–438. J. Harrington and M. Tabain (Psychology Press, New York), pp. Westbury, J. R. (1979). “Aspects of the temporal control of voicing in con- 275–297. sonant clusters in English,” Texas Linguist. Forum 14, 1–304. Ruty, N., Pelorson, X., and Van Hirtum, A. (2008). “Influence of acoustic Whalen, D. H., and Levitt, A. G. (1995). “The universality of intrinsic F0 of waveguides lengths on self-sustained oscillations: Theoretical prediction vowels,” J. Phonetics 23, 349–366. and experimental validation,” J. Acoust. Soc. Am. 123, 3121. Yoshioka, H., Lo¨fqvist, A., and Hirose, H. (1981). “Laryngeal adjustments Sivasankar, M., Erickson, E., Schneider, S., and Hawes, A. (2008). in the production of consonant clusters and geminates in American Eng- “Phonatory effects of airway dehydration: Preliminary evidence for lish,” J. Acoust. Soc. Am. 70, 1615–1623.

3244 J. Acoust. Soc. Am., Vol. 129, No. 5, May 2011 Koenig et al.: Phonemic effects on voicing thresholds

Downloaded 11 May 2011 to 128.36.197.66. Redistribution subject to ASA license or copyright; see http://asadl.org/journals/doc/ASALIB-home/info/terms.jsp